Re: [PATCH/RFC RESEND] leds: Use set_brightness_work for brightness_set ops that can sleep

[Pavel Machek]

On Wed 2015-07-01 12:47:02, Jacek Anaszewski wrote:
> On 07/01/2015 09:43 AM, Pavel Machek wrote:
> >On Wed 2015-07-01 09:28:52, Jacek Anaszewski wrote:
> >>On 06/30/2015 07:46 PM, Pavel Machek wrote:
> >>>On Tue 2015-06-30 15:06:19, Jacek Anaszewski wrote:
> >>>>On 06/30/2015 01:58 PM, Pavel Machek wrote:
> >>>>>On Tue 2015-06-30 10:01:08, Jacek Anaszewski wrote:
> >>>>>>This patch rearranges the core LED subsystem code, so that it
> >>>>>>now removes from drivers the responsibility of using work queues
> >>>>>>internally in case their brightness_set ops can sleep.
> >>>>>>Addition of two flags: LED_BRIGHTNESS_FAST and LED_BLINK_DISABLE
> >>>>>>as well as new_brightness_value property to the struct led_classdev
> >>>>>>allows for employing existing set_brightness_work to do the job.
> >>>>>>The modifications allow also to get rid of brightness_set_sync op,
> >>>>>>as flash LED devices can now be handled properly only basing on the
> >>>>>>SET_BRIGHTNESS_SYNC flag.
> >>>>>
> >>>>>Are you sure this is good idea?
> >>>>>
> >>>>>You'll now use single callback for blocking and non-blocking
> >>>>>behaviour. I'm pretty sure stuff like lockdep will have some fun with
> >>>>>that.
> >>>>
> >>>>I enabled "Lock Debugging" options and didn't get any warning.
> >>>>Could you describe the use case you are thinking of?
> >>>
> >>>You may get one when one of the sleeping functions uses some lock...
> >>
> >>Drivers which use spin_lock in their brightness_set op will have to
> >>set LED_BRIGHTNESS_FAST flag, which will instruct the LED core to
> >>call the op synchronously. On the other hand drivers which can sleep
> >>in their brightness_set op won't set the flag, which will make LED core
> >>delegating the op to the work queue task. It is also possible that
> >>driver with brightness_set op that can sleep set SET_BRIGHTNESS_SYNC
> >>flag - then LED core will call it in a synchronous way from
> >>led_brightness_set and it will schedule work queue task in case
> >>the op is called from triggers.
> >
> >I understand this "works".
> >
> >>If you want to NAK the patch, please come up with detailed analysis
> >>on how it can cause problems. Without this I infer that you didn't
> >>spend a second on analyzing the code. This is counterproductive.
> >
> >NAK.
> >
> >Because calling two functions with different semantics through same
> >function pointer is extremely ugly, and _will_ cause lockdep
> >problems. Talk to the lockdep people for details.
> 
> Which two functions are you thinking of? There is a single
> brightness_set op to call.

Yes, and brightness_set may or may not sleep according to a flag
somewhere. Just use two function pointers, one of them will be always
NULL. You can keep the flag if you want to.

									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/